Weekly Communication

Week May31, 2002

Jin: . Due to the need to investgate the performance of M3DO, I found these 
       Tools on IBM SP

       1)ZEROFAULT to detect any memory problem
         - module load ZeroFault
         - compile M3DP with -g option
         - run m3dp.x with -zf -u option

       2)HPMCOUNT to moniter performance of the whole code
         poe+

       3)HPMVIZ to moniter performance on segment of the code
         - add 
            #include "f_hpm.h"
             call f_hpminit(taskID, "weqn --> lowpois")
             call f_hpmstart(1,"weqn --> lowpois")
                  ..... code segment .....
             call f_hpmstop(1)
             call f_hpmterminate(taskID)
           to the segment of code needed to be monitered
         - compile and link with $HPMTOOLKIT
           ex: seaborg/m3dploki/mhd/driver/Makefile
                                m3d/code/Makefile
                                         m1.F/subroutine weqn(...)

       4)VAMPIR to moniter data communication and computation performance
         in parallel enviroment
         - module load vampir
         - compile and link M3DP with $VTLIB 

     . w3.pppl.gov/ion~jchen/doc/KSP-PC.html
       - contacted PETSc people regarding Krylov Subspace method with 
         Preconditioner / Direct Sequential Method
       - experimented available KSPs and proconditioners, 
         GMRES with ILU preconditioning seems to be the most suitable one for M3DP

     . communicated with ODU
       w3.pppl.gov/~jchen/collaboration.html

     . spectral element: reading papers

Week May24, 2002

Jin: . back to work from May 21
     . preparing the SIAM annual meeting presentation
     . showed the 1st presentation draft to Steve on May 22.
     . docs for grid/mesh/matrix were added at w3.pppl.gov/~jchen/doc.html
     . having a better understanding for the matrix structure 
       solved in m3dp
     . making the grid/mesh part be independent for spectral element 
       implementation
     . making the finite element part be independent for finite element
       order improvement
     . making the matrix part be independent for tuning between different 
       solvers and preconditioner

ODU: Florin is trying to switch from petsc-1.2.0 to 1.2.2 
     in order to take advantage of hypre

RPI: Katia is working on the 2D model problem